drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "osd.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 /*
  50  * This request ties the struct request and struct
  51  * blkvsc_request/hv_storvsc_request together A struct request may be
  52  * represented by 1 or more struct blkvsc_request
  53  */
  54 struct blkvsc_request_group {
  55         int outstanding;
  56         int status;
  57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  58 };
  59
  60 struct blkvsc_request {
  61         /* blkvsc_request_group.blkvsc_req_list */
  62         struct list_head req_entry;
  63
  64         /* block_device_context.pending_list */
  65         struct list_head pend_entry;
  66
  67         /* This may be null if we generate a request internally */
  68         struct request *req;
  69
  70         struct block_device_context *dev;
  71
  72         /* The group this request is part of. Maybe null */
  73         struct blkvsc_request_group *group;
  74
  75         wait_queue_head_t wevent;
  76         int cond;
  77
  78         int write;
  79         sector_t sector_start;
  80         unsigned long sector_count;
  81
  82         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  83         unsigned char cmd_len;
  84         unsigned char cmnd[MAX_COMMAND_SIZE];
  85
  86         struct hv_storvsc_request request;
  87         /*
  88          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  89          * because - The extension buffer falls right here and is pointed to by
  90          * request.Extension;
  91          * Which sounds like a horrible idea, who designed this?
  92          */
  93 };
  94
  95 /* Per device structure */
  96 struct block_device_context {
  97         /* point back to our device context */
  98         struct vm_device *device_ctx;
  99         struct kmem_cache *request_pool;
 100         spinlock_t lock;
 101         struct gendisk *gd;
 102         enum blkvsc_device_type device_type;
 103         struct list_head pending_list;
 104
 105         unsigned char device_id[64];
 106         unsigned int device_id_len;
 107         int num_outstanding_reqs;
 108         int shutting_down;
 109         int media_not_present;
 110         unsigned int sector_size;
 111         sector_t capacity;
 112         unsigned int port;
 113         unsigned char path;
 114         unsigned char target;
 115         int users;
 116 };
 117
 118 /* Per driver */
 119 struct blkvsc_driver_context {
 120         /* !! These must be the first 2 fields !! */
 121         /* FIXME this is a bug! */
 122         struct driver_context drv_ctx;
 123         struct storvsc_driver_object drv_obj;
 124 };
 125
 126 /* Static decl */
 127 static DEFINE_MUTEX(blkvsc_mutex);
 128 static int blkvsc_probe(struct device *dev);
 129 static int blkvsc_remove(struct device *device);
 130 static void blkvsc_shutdown(struct device *device);
 131
 132 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 133 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 134 static int blkvsc_media_changed(struct gendisk *gd);
 135 static int blkvsc_revalidate_disk(struct gendisk *gd);
 136 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 137 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 138                         unsigned cmd, unsigned long argument);
 139 static void blkvsc_request(struct request_queue *queue);
 140 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 141 static int blkvsc_do_request(struct block_device_context *blkdev,
 142                              struct request *req);
 143 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 144                 void (*request_completion)(struct hv_storvsc_request *));
 145 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 146 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 147 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 148 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 149 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 150 static int blkvsc_do_flush(struct block_device_context *blkdev);
 151 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 152 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 153
 154 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 155 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 156 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 157
 158 /* The one and only one */
 159 static struct blkvsc_driver_context g_blkvsc_drv;
 160
 161 static const struct block_device_operations block_ops = {
 162         .owner = THIS_MODULE,
 163         .open = blkvsc_open,
 164         .release = blkvsc_release,
 165         .media_changed = blkvsc_media_changed,
 166         .revalidate_disk = blkvsc_revalidate_disk,
 167         .getgeo = blkvsc_getgeo,
 168         .ioctl  = blkvsc_ioctl,
 169 };
 170
 171 /*
 172  * blkvsc_drv_init -  BlkVsc driver initialization.
 173  */
 174 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 175 {
 176         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 177         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 178         int ret;
 179
 180         vmbus_get_interface(&storvsc_drv_obj->Base.VmbusChannelInterface);
 181
 182         storvsc_drv_obj->RingBufferSize = blkvsc_ringbuffer_size;
 183
 184         /* Callback to client driver to complete the initialization */
 185         drv_init(&storvsc_drv_obj->Base);
 186
 187         drv_ctx->driver.name = storvsc_drv_obj->Base.name;
 188         memcpy(&drv_ctx->class_id, &storvsc_drv_obj->Base.deviceType,
 189                sizeof(struct hv_guid));
 190
 191         drv_ctx->probe = blkvsc_probe;
 192         drv_ctx->remove = blkvsc_remove;
 193         drv_ctx->shutdown = blkvsc_shutdown;
 194
 195         /* The driver belongs to vmbus */
 196         ret = vmbus_child_driver_register(drv_ctx);
 197
 198         return ret;
 199 }
 200
 201 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 202 {
 203         struct device **curr = (struct device **)data;
 204         *curr = dev;
 205         return 1; /* stop iterating */
 206 }
 207
 208 static void blkvsc_drv_exit(void)
 209 {
 210         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 211         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 212         struct device *current_dev;
 213         int ret;
 214
 215         while (1) {
 216                 current_dev = NULL;
 217
 218                 /* Get the device */
 219                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
 220                                              (void *) &current_dev,
 221                                              blkvsc_drv_exit_cb);
 222
 223                 if (ret)
 224                         DPRINT_WARN(BLKVSC_DRV,
 225                                     "driver_for_each_device returned %d", ret);
 226
 227
 228                 if (current_dev == NULL)
 229                         break;
 230
 231                 /* Initiate removal from the top-down */
 232                 device_unregister(current_dev);
 233         }
 234
 235         if (storvsc_drv_obj->Base.OnCleanup)
 236                 storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
 237
 238         vmbus_child_driver_unregister(drv_ctx);
 239
 240         return;
 241 }
 242
 243 /*
 244  * blkvsc_probe - Add a new device for this driver
 245  */
 246 static int blkvsc_probe(struct device *device)
 247 {
 248         struct driver_context *driver_ctx =
 249                                 driver_to_driver_context(device->driver);
 250         struct blkvsc_driver_context *blkvsc_drv_ctx =
 251                                 (struct blkvsc_driver_context *)driver_ctx;
 252         struct storvsc_driver_object *storvsc_drv_obj =
 253                                 &blkvsc_drv_ctx->drv_obj;
 254         struct vm_device *device_ctx = device_to_vm_device(device);
 255         struct hv_device *device_obj = &device_ctx->device_obj;
 256
 257         struct block_device_context *blkdev = NULL;
 258         struct storvsc_device_info device_info;
 259         int major = 0;
 260         int devnum = 0;
 261         int ret = 0;
 262         static int ide0_registered;
 263         static int ide1_registered;
 264
 265         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 266
 267         if (!storvsc_drv_obj->Base.OnDeviceAdd) {
 268                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 269                 ret = -1;
 270                 goto Cleanup;
 271         }
 272
 273         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 274         if (!blkdev) {
 275                 ret = -ENOMEM;
 276                 goto Cleanup;
 277         }
 278
 279         INIT_LIST_HEAD(&blkdev->pending_list);
 280
 281         /* Initialize what we can here */
 282         spin_lock_init(&blkdev->lock);
 283
 284         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 285         /*      sizeof(struct blkvsc_request)); */
 286
 287         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
 288                                         sizeof(struct blkvsc_request) +
 289                                         storvsc_drv_obj->RequestExtSize, 0,
 290                                         SLAB_HWCACHE_ALIGN, NULL);
 291         if (!blkdev->request_pool) {
 292                 ret = -ENOMEM;
 293                 goto Cleanup;
 294         }
 295
 296
 297         /* Call to the vsc driver to add the device */
 298         ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
 299         if (ret != 0) {
 300                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 301                 goto Cleanup;
 302         }
 303
 304         blkdev->device_ctx = device_ctx;
 305         /* this identified the device 0 or 1 */
 306         blkdev->target = device_info.TargetId;
 307         /* this identified the ide ctrl 0 or 1 */
 308         blkdev->path = device_info.PathId;
 309
 310         dev_set_drvdata(device, blkdev);
 311
 312         /* Calculate the major and device num */
 313         if (blkdev->path == 0) {
 314                 major = IDE0_MAJOR;
 315                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 316
 317                 if (!ide0_registered) {
 318                         ret = register_blkdev(major, "ide");
 319                         if (ret != 0) {
 320                                 DPRINT_ERR(BLKVSC_DRV,
 321                                            "register_blkdev() failed! ret %d",
 322                                            ret);
 323                                 goto Remove;
 324                         }
 325
 326                         ide0_registered = 1;
 327                 }
 328         } else if (blkdev->path == 1) {
 329                 major = IDE1_MAJOR;
 330                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 331
 332                 if (!ide1_registered) {
 333                         ret = register_blkdev(major, "ide");
 334                         if (ret != 0) {
 335                                 DPRINT_ERR(BLKVSC_DRV,
 336                                            "register_blkdev() failed! ret %d",
 337                                            ret);
 338                                 goto Remove;
 339                         }
 340
 341                         ide1_registered = 1;
 342                 }
 343         } else {
 344                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 345                 ret = -1;
 346                 goto Cleanup;
 347         }
 348
 349         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 350
 351         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 352         if (!blkdev->gd) {
 353                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 354                 ret = -1;
 355                 goto Cleanup;
 356         }
 357
 358         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 359
 360         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 361         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 362         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 363         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 364         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 365
 366         blkdev->gd->major = major;
 367         if (devnum == 1 || devnum == 3)
 368                 blkdev->gd->first_minor = BLKVSC_MINORS;
 369         else
 370                 blkdev->gd->first_minor = 0;
 371         blkdev->gd->fops = &block_ops;
 372         blkdev->gd->private_data = blkdev;
 373         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 374
 375         blkvsc_do_inquiry(blkdev);
 376         if (blkdev->device_type == DVD_TYPE) {
 377                 set_disk_ro(blkdev->gd, 1);
 378                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 379                 blkvsc_do_read_capacity(blkdev);
 380         } else {
 381                 blkvsc_do_read_capacity16(blkdev);
 382         }
 383
 384         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 385         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 386         /* go! */
 387         add_disk(blkdev->gd);
 388
 389         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 390                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 391                     blkdev->sector_size);
 392
 393         return ret;
 394
 395 Remove:
 396         storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 397
 398 Cleanup:
 399         if (blkdev) {
 400                 if (blkdev->request_pool) {
 401                         kmem_cache_destroy(blkdev->request_pool);
 402                         blkdev->request_pool = NULL;
 403                 }
 404                 kfree(blkdev);
 405                 blkdev = NULL;
 406         }
 407
 408         return ret;
 409 }
 410
 411 static void blkvsc_shutdown(struct device *device)
 412 {
 413         struct block_device_context *blkdev = dev_get_drvdata(device);
 414         unsigned long flags;
 415
 416         if (!blkdev)
 417                 return;
 418
 419         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 420                    blkdev->users, blkdev->gd->disk_name);
 421
 422         spin_lock_irqsave(&blkdev->lock, flags);
 423
 424         blkdev->shutting_down = 1;
 425
 426         blk_stop_queue(blkdev->gd->queue);
 427
 428         spin_unlock_irqrestore(&blkdev->lock, flags);
 429
 430         while (blkdev->num_outstanding_reqs) {
 431                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 432                             blkdev->num_outstanding_reqs);
 433                 udelay(100);
 434         }
 435
 436         blkvsc_do_flush(blkdev);
 437
 438         spin_lock_irqsave(&blkdev->lock, flags);
 439
 440         blkvsc_cancel_pending_reqs(blkdev);
 441
 442         spin_unlock_irqrestore(&blkdev->lock, flags);
 443 }
 444
 445 static int blkvsc_do_flush(struct block_device_context *blkdev)
 446 {
 447         struct blkvsc_request *blkvsc_req;
 448
 449         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 450
 451         if (blkdev->device_type != HARDDISK_TYPE)
 452                 return 0;
 453
 454         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 455         if (!blkvsc_req)
 456                 return -ENOMEM;
 457
 458         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 459         init_waitqueue_head(&blkvsc_req->wevent);
 460         blkvsc_req->dev = blkdev;
 461         blkvsc_req->req = NULL;
 462         blkvsc_req->write = 0;
 463
 464         blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
 465         blkvsc_req->request.DataBuffer.Offset = 0;
 466         blkvsc_req->request.DataBuffer.Length = 0;
 467
 468         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 469         blkvsc_req->cmd_len = 10;
 470
 471         /*
 472          * Set this here since the completion routine may be invoked and
 473          * completed before we return
 474          */
 475         blkvsc_req->cond = 0;
 476         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 477
 478         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 479
 480         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 481
 482         return 0;
 483 }
 484
 485 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 486 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 487 {
 488         struct blkvsc_request *blkvsc_req;
 489         struct page *page_buf;
 490         unsigned char *buf;
 491         unsigned char device_type;
 492
 493         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 494
 495         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 496         if (!blkvsc_req)
 497                 return -ENOMEM;
 498
 499         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 500         page_buf = alloc_page(GFP_KERNEL);
 501         if (!page_buf) {
 502                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 503                 return -ENOMEM;
 504         }
 505
 506         init_waitqueue_head(&blkvsc_req->wevent);
 507         blkvsc_req->dev = blkdev;
 508         blkvsc_req->req = NULL;
 509         blkvsc_req->write = 0;
 510
 511         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 512         blkvsc_req->request.DataBuffer.Offset = 0;
 513         blkvsc_req->request.DataBuffer.Length = 64;
 514
 515         blkvsc_req->cmnd[0] = INQUIRY;
 516         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 517         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 518         blkvsc_req->cmnd[4] = 64;
 519         blkvsc_req->cmd_len = 6;
 520
 521         /*
 522          * Set this here since the completion routine may be invoked and
 523          * completed before we return
 524          */
 525         blkvsc_req->cond = 0;
 526
 527         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 528
 529         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 530                    blkvsc_req, blkvsc_req->cond);
 531
 532         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 533
 534         buf = kmap(page_buf);
 535
 536         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 537         /* be to le */
 538         device_type = buf[0] & 0x1F;
 539
 540         if (device_type == 0x0) {
 541                 blkdev->device_type = HARDDISK_TYPE;
 542         } else if (device_type == 0x5) {
 543                 blkdev->device_type = DVD_TYPE;
 544         } else {
 545                 /* TODO: this is currently unsupported device type */
 546                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 547         }
 548
 549         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 550
 551         blkdev->device_id_len = buf[7];
 552         if (blkdev->device_id_len > 64)
 553                 blkdev->device_id_len = 64;
 554
 555         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 556         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 557          * blkdev->device_id_len); */
 558
 559         kunmap(page_buf);
 560
 561         __free_page(page_buf);
 562
 563         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 564
 565         return 0;
 566 }
 567
 568 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 569 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 570 {
 571         struct blkvsc_request *blkvsc_req;
 572         struct page *page_buf;
 573         unsigned char *buf;
 574         struct scsi_sense_hdr sense_hdr;
 575
 576         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 577
 578         blkdev->sector_size = 0;
 579         blkdev->capacity = 0;
 580         blkdev->media_not_present = 0; /* assume a disk is present */
 581
 582         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 583         if (!blkvsc_req)
 584                 return -ENOMEM;
 585
 586         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 587         page_buf = alloc_page(GFP_KERNEL);
 588         if (!page_buf) {
 589                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 590                 return -ENOMEM;
 591         }
 592
 593         init_waitqueue_head(&blkvsc_req->wevent);
 594         blkvsc_req->dev = blkdev;
 595         blkvsc_req->req = NULL;
 596         blkvsc_req->write = 0;
 597
 598         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 599         blkvsc_req->request.DataBuffer.Offset = 0;
 600         blkvsc_req->request.DataBuffer.Length = 8;
 601
 602         blkvsc_req->cmnd[0] = READ_CAPACITY;
 603         blkvsc_req->cmd_len = 16;
 604
 605         /*
 606          * Set this here since the completion routine may be invoked
 607          * and completed before we return
 608          */
 609         blkvsc_req->cond = 0;
 610
 611         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 612
 613         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 614                    blkvsc_req, blkvsc_req->cond);
 615
 616         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 617
 618         /* check error */
 619         if (blkvsc_req->request.Status) {
 620                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 621                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 622
 623                 if (sense_hdr.asc == 0x3A) {
 624                         /* Medium not present */
 625                         blkdev->media_not_present = 1;
 626                 }
 627                 return 0;
 628         }
 629         buf = kmap(page_buf);
 630
 631         /* be to le */
 632         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 633                             (buf[2] << 8) | buf[3]) + 1;
 634         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 635                               (buf[6] << 8) | buf[7];
 636
 637         kunmap(page_buf);
 638
 639         __free_page(page_buf);
 640
 641         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 642
 643         return 0;
 644 }
 645
 646 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 647 {
 648         struct blkvsc_request *blkvsc_req;
 649         struct page *page_buf;
 650         unsigned char *buf;
 651         struct scsi_sense_hdr sense_hdr;
 652
 653         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 654
 655         blkdev->sector_size = 0;
 656         blkdev->capacity = 0;
 657         blkdev->media_not_present = 0; /* assume a disk is present */
 658
 659         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 660         if (!blkvsc_req)
 661                 return -ENOMEM;
 662
 663         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 664         page_buf = alloc_page(GFP_KERNEL);
 665         if (!page_buf) {
 666                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 667                 return -ENOMEM;
 668         }
 669
 670         init_waitqueue_head(&blkvsc_req->wevent);
 671         blkvsc_req->dev = blkdev;
 672         blkvsc_req->req = NULL;
 673         blkvsc_req->write = 0;
 674
 675         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 676         blkvsc_req->request.DataBuffer.Offset = 0;
 677         blkvsc_req->request.DataBuffer.Length = 12;
 678
 679         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 680         blkvsc_req->cmd_len = 16;
 681
 682         /*
 683          * Set this here since the completion routine may be invoked
 684          * and completed before we return
 685          */
 686         blkvsc_req->cond = 0;
 687
 688         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 689
 690         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 691                    blkvsc_req, blkvsc_req->cond);
 692
 693         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 694
 695         /* check error */
 696         if (blkvsc_req->request.Status) {
 697                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 698                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 699                 if (sense_hdr.asc == 0x3A) {
 700                         /* Medium not present */
 701                         blkdev->media_not_present = 1;
 702                 }
 703                 return 0;
 704         }
 705         buf = kmap(page_buf);
 706
 707         /* be to le */
 708         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 709         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 710
 711 #if 0
 712         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 713                             (buf[2] << 8) | buf[3]) + 1;
 714         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 715                               (buf[6] << 8) | buf[7];
 716 #endif
 717
 718         kunmap(page_buf);
 719
 720         __free_page(page_buf);
 721
 722         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 723
 724         return 0;
 725 }
 726
 727 /*
 728  * blkvsc_remove() - Callback when our device is removed
 729  */
 730 static int blkvsc_remove(struct device *device)
 731 {
 732         struct driver_context *driver_ctx =
 733                                 driver_to_driver_context(device->driver);
 734         struct blkvsc_driver_context *blkvsc_drv_ctx =
 735                                 (struct blkvsc_driver_context *)driver_ctx;
 736         struct storvsc_driver_object *storvsc_drv_obj =
 737                                 &blkvsc_drv_ctx->drv_obj;
 738         struct vm_device *device_ctx = device_to_vm_device(device);
 739         struct hv_device *device_obj = &device_ctx->device_obj;
 740         struct block_device_context *blkdev = dev_get_drvdata(device);
 741         unsigned long flags;
 742         int ret;
 743
 744         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 745
 746         if (!storvsc_drv_obj->Base.OnDeviceRemove)
 747                 return -1;
 748
 749         /*
 750          * Call to the vsc driver to let it know that the device is being
 751          * removed
 752          */
 753         ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 754         if (ret != 0) {
 755                 /* TODO: */
 756                 DPRINT_ERR(BLKVSC_DRV,
 757                            "unable to remove blkvsc device (ret %d)", ret);
 758         }
 759
 760         /* Get to a known state */
 761         spin_lock_irqsave(&blkdev->lock, flags);
 762
 763         blkdev->shutting_down = 1;
 764
 765         blk_stop_queue(blkdev->gd->queue);
 766
 767         spin_unlock_irqrestore(&blkdev->lock, flags);
 768
 769         while (blkdev->num_outstanding_reqs) {
 770                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 771                             blkdev->num_outstanding_reqs);
 772                 udelay(100);
 773         }
 774
 775         blkvsc_do_flush(blkdev);
 776
 777         spin_lock_irqsave(&blkdev->lock, flags);
 778
 779         blkvsc_cancel_pending_reqs(blkdev);
 780
 781         spin_unlock_irqrestore(&blkdev->lock, flags);
 782
 783         blk_cleanup_queue(blkdev->gd->queue);
 784
 785         del_gendisk(blkdev->gd);
 786
 787         kmem_cache_destroy(blkdev->request_pool);
 788
 789         kfree(blkdev);
 790
 791         return ret;
 792 }
 793
 794 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 795 {
 796         /* ASSERT(blkvsc_req->req); */
 797         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 798
 799         blkvsc_req->cmd_len = 16;
 800
 801         if (blkvsc_req->sector_start > 0xffffffff) {
 802                 if (rq_data_dir(blkvsc_req->req)) {
 803                         blkvsc_req->write = 1;
 804                         blkvsc_req->cmnd[0] = WRITE_16;
 805                 } else {
 806                         blkvsc_req->write = 0;
 807                         blkvsc_req->cmnd[0] = READ_16;
 808                 }
 809
 810                 blkvsc_req->cmnd[1] |=
 811                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 812
 813                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 814                                 cpu_to_be64(blkvsc_req->sector_start);
 815                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 816                                 cpu_to_be32(blkvsc_req->sector_count);
 817         } else if ((blkvsc_req->sector_count > 0xff) ||
 818                    (blkvsc_req->sector_start > 0x1fffff)) {
 819                 if (rq_data_dir(blkvsc_req->req)) {
 820                         blkvsc_req->write = 1;
 821                         blkvsc_req->cmnd[0] = WRITE_10;
 822                 } else {
 823                         blkvsc_req->write = 0;
 824                         blkvsc_req->cmnd[0] = READ_10;
 825                 }
 826
 827                 blkvsc_req->cmnd[1] |=
 828                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 829
 830                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 831                                 cpu_to_be32(blkvsc_req->sector_start);
 832                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 833                                 cpu_to_be16(blkvsc_req->sector_count);
 834         } else {
 835                 if (rq_data_dir(blkvsc_req->req)) {
 836                         blkvsc_req->write = 1;
 837                         blkvsc_req->cmnd[0] = WRITE_6;
 838                 } else {
 839                         blkvsc_req->write = 0;
 840                         blkvsc_req->cmnd[0] = READ_6;
 841                 }
 842
 843                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 844                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 845                 blkvsc_req->cmnd[1] &= 0x1f;
 846                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 847         }
 848 }
 849
 850 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 851                         void (*request_completion)(struct hv_storvsc_request *))
 852 {
 853         struct block_device_context *blkdev = blkvsc_req->dev;
 854         struct vm_device *device_ctx = blkdev->device_ctx;
 855         struct driver_context *driver_ctx =
 856                         driver_to_driver_context(device_ctx->device.driver);
 857         struct blkvsc_driver_context *blkvsc_drv_ctx =
 858                         (struct blkvsc_driver_context *)driver_ctx;
 859         struct storvsc_driver_object *storvsc_drv_obj =
 860                         &blkvsc_drv_ctx->drv_obj;
 861         struct hv_storvsc_request *storvsc_req;
 862         int ret;
 863
 864         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 865                    "req %p type %s start_sector %lu count %ld offset %d "
 866                    "len %d\n", blkvsc_req,
 867                    (blkvsc_req->write) ? "WRITE" : "READ",
 868                    (unsigned long) blkvsc_req->sector_start,
 869                    blkvsc_req->sector_count,
 870                    blkvsc_req->request.DataBuffer.Offset,
 871                    blkvsc_req->request.DataBuffer.Length);
 872 #if 0
 873         for (i = 0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++) {
 874                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 875                            "req %p pfn[%d] %llx\n",
 876                            blkvsc_req, i,
 877                            blkvsc_req->request.DataBuffer.PfnArray[i]);
 878         }
 879 #endif
 880
 881         storvsc_req = &blkvsc_req->request;
 882         storvsc_req->Extension = (void *)((unsigned long)blkvsc_req +
 883                                           sizeof(struct blkvsc_request));
 884
 885         storvsc_req->Type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 886
 887         storvsc_req->OnIOCompletion = request_completion;
 888         storvsc_req->Context = blkvsc_req;
 889
 890         storvsc_req->Host = blkdev->port;
 891         storvsc_req->Bus = blkdev->path;
 892         storvsc_req->TargetId = blkdev->target;
 893         storvsc_req->LunId = 0;  /* this is not really used at all */
 894
 895         storvsc_req->CdbLen = blkvsc_req->cmd_len;
 896         storvsc_req->Cdb = blkvsc_req->cmnd;
 897
 898         storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
 899         storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
 900
 901         ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj,
 902                                            &blkvsc_req->request);
 903         if (ret == 0)
 904                 blkdev->num_outstanding_reqs++;
 905
 906         return ret;
 907 }
 908
 909 /*
 910  * We break the request into 1 or more blkvsc_requests and submit
 911  * them.  If we cant submit them all, we put them on the
 912  * pending_list. The blkvsc_request() will work on the pending_list.
 913  */
 914 static int blkvsc_do_request(struct block_device_context *blkdev,
 915                              struct request *req)
 916 {
 917         struct bio *bio = NULL;
 918         struct bio_vec *bvec = NULL;
 919         struct bio_vec *prev_bvec = NULL;
 920         struct blkvsc_request *blkvsc_req = NULL;
 921         struct blkvsc_request *tmp;
 922         int databuf_idx = 0;
 923         int seg_idx = 0;
 924         sector_t start_sector;
 925         unsigned long num_sectors = 0;
 926         int ret = 0;
 927         int pending = 0;
 928         struct blkvsc_request_group *group = NULL;
 929
 930         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 931                   (unsigned long)blk_rq_pos(req));
 932
 933         /* Create a group to tie req to list of blkvsc_reqs */
 934         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 935         if (!group)
 936                 return -ENOMEM;
 937
 938         INIT_LIST_HEAD(&group->blkvsc_req_list);
 939         group->outstanding = group->status = 0;
 940
 941         start_sector = blk_rq_pos(req);
 942
 943         /* foreach bio in the request */
 944         if (req->bio) {
 945                 for (bio = req->bio; bio; bio = bio->bi_next) {
 946                         /*
 947                          * Map this bio into an existing or new storvsc request
 948                          */
 949                         bio_for_each_segment(bvec, bio, seg_idx) {
 950                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 951                                            "- req %p bio %p bvec %p seg_idx %d "
 952                                            "databuf_idx %d\n", req, bio, bvec,
 953                                            seg_idx, databuf_idx);
 954
 955                                 /* Get a new storvsc request */
 956                                 /* 1st-time */
 957                                 if ((!blkvsc_req) ||
 958                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 959                                     /* hole at the begin of page */
 960                                     || (bvec->bv_offset != 0) ||
 961                                     /* hold at the end of page */
 962                                     (prev_bvec &&
 963                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 964                                         /* submit the prev one */
 965                                         if (blkvsc_req) {
 966                                                 blkvsc_req->sector_start = start_sector;
 967                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 968
 969                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 970                                                 blkvsc_init_rw(blkvsc_req);
 971                                         }
 972
 973                                         /*
 974                                          * Create new blkvsc_req to represent
 975                                          * the current bvec
 976                                          */
 977                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 978                                         if (!blkvsc_req) {
 979                                                 /* free up everything */
 980                                                 list_for_each_entry_safe(
 981                                                         blkvsc_req, tmp,
 982                                                         &group->blkvsc_req_list,
 983                                                         req_entry) {
 984                                                         list_del(&blkvsc_req->req_entry);
 985                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 986                                                 }
 987
 988                                                 kmem_cache_free(blkdev->request_pool, group);
 989                                                 return -ENOMEM;
 990                                         }
 991
 992                                         memset(blkvsc_req, 0,
 993                                                sizeof(struct blkvsc_request));
 994
 995                                         blkvsc_req->dev = blkdev;
 996                                         blkvsc_req->req = req;
 997                                         blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
 998                                         blkvsc_req->request.DataBuffer.Length = 0;
 999
1000                                         /* Add to the group */
1001                                         blkvsc_req->group = group;
1002                                         blkvsc_req->group->outstanding++;
1003                                         list_add_tail(&blkvsc_req->req_entry,
1004                                                 &blkvsc_req->group->blkvsc_req_list);
1005
1006                                         start_sector += num_sectors;
1007                                         num_sectors = 0;
1008                                         databuf_idx = 0;
1009                                 }
1010
1011                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1012                                 blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
1013                                 blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
1014
1015                                 prev_bvec = bvec;
1016
1017                                 databuf_idx++;
1018                                 num_sectors += bvec->bv_len >> 9;
1019
1020                         } /* bio_for_each_segment */
1021
1022                 } /* rq_for_each_bio */
1023         }
1024
1025         /* Handle the last one */
1026         if (blkvsc_req) {
1027                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1028                            blkdev, req, blkvsc_req->group,
1029                            blkvsc_req->group->outstanding);
1030
1031                 blkvsc_req->sector_start = start_sector;
1032                 sector_div(blkvsc_req->sector_start,
1033                            (blkdev->sector_size >> 9));
1034
1035                 blkvsc_req->sector_count = num_sectors /
1036                                            (blkdev->sector_size >> 9);
1037
1038                 blkvsc_init_rw(blkvsc_req);
1039         }
1040
1041         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1042                 if (pending) {
1043                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1044                                    "pending_list - blkvsc_req %p start_sect %lu"
1045                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1046                                    (unsigned long)blkvsc_req->sector_start,
1047                                    blkvsc_req->sector_count,
1048                                    (unsigned long)start_sector,
1049                                    (unsigned long)num_sectors);
1050
1051                         list_add_tail(&blkvsc_req->pend_entry,
1052                                       &blkdev->pending_list);
1053                 } else {
1054                         ret = blkvsc_submit_request(blkvsc_req,
1055                                                     blkvsc_request_completion);
1056                         if (ret == -1) {
1057                                 pending = 1;
1058                                 list_add_tail(&blkvsc_req->pend_entry,
1059                                               &blkdev->pending_list);
1060                         }
1061
1062                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1063                                    "start_sect %lu sect_count %ld (%lu %ld) "
1064                                    "ret %d\n", blkvsc_req,
1065                                    (unsigned long)blkvsc_req->sector_start,
1066                                    blkvsc_req->sector_count,
1067                                    (unsigned long)start_sector,
1068                                    num_sectors, ret);
1069                 }
1070         }
1071
1072         return pending;
1073 }
1074
1075 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1076 {
1077         struct blkvsc_request *blkvsc_req =
1078                         (struct blkvsc_request *)request->Context;
1079         struct block_device_context *blkdev =
1080                         (struct block_device_context *)blkvsc_req->dev;
1081         struct scsi_sense_hdr sense_hdr;
1082
1083         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1084                    blkvsc_req);
1085
1086         blkdev->num_outstanding_reqs--;
1087
1088         if (blkvsc_req->request.Status)
1089                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1090                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1091                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1092
1093         blkvsc_req->cond = 1;
1094         wake_up_interruptible(&blkvsc_req->wevent);
1095 }
1096
1097 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1098 {
1099         struct blkvsc_request *blkvsc_req =
1100                         (struct blkvsc_request *)request->Context;
1101         struct block_device_context *blkdev =
1102                         (struct block_device_context *)blkvsc_req->dev;
1103         unsigned long flags;
1104         struct blkvsc_request *comp_req, *tmp;
1105
1106         /* ASSERT(blkvsc_req->group); */
1107
1108         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1109                    "sect_start %lu sect_count %ld len %d group outstd %d "
1110                    "total outstd %d\n",
1111                    blkdev, blkvsc_req, blkvsc_req->group,
1112                    (blkvsc_req->write) ? "WRITE" : "READ",
1113                    (unsigned long)blkvsc_req->sector_start,
1114                    blkvsc_req->sector_count,
1115                    blkvsc_req->request.DataBuffer.Length,
1116                    blkvsc_req->group->outstanding,
1117                    blkdev->num_outstanding_reqs);
1118
1119         spin_lock_irqsave(&blkdev->lock, flags);
1120
1121         blkdev->num_outstanding_reqs--;
1122         blkvsc_req->group->outstanding--;
1123
1124         /*
1125          * Only start processing when all the blkvsc_reqs are
1126          * completed. This guarantees no out-of-order blkvsc_req
1127          * completion when calling end_that_request_first()
1128          */
1129         if (blkvsc_req->group->outstanding == 0) {
1130                 list_for_each_entry_safe(comp_req, tmp,
1131                                          &blkvsc_req->group->blkvsc_req_list,
1132                                          req_entry) {
1133                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1134                                    "sect_start %lu sect_count %ld\n",
1135                                    comp_req,
1136                                    (unsigned long)comp_req->sector_start,
1137                                    comp_req->sector_count);
1138
1139                         list_del(&comp_req->req_entry);
1140
1141                         if (!__blk_end_request(comp_req->req,
1142                                 (!comp_req->request.Status ? 0 : -EIO),
1143                                 comp_req->sector_count * blkdev->sector_size)) {
1144                                 /*
1145                                  * All the sectors have been xferred ie the
1146                                  * request is done
1147                                  */
1148                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1149                                            comp_req->req);
1150                                 kmem_cache_free(blkdev->request_pool,
1151                                                 comp_req->group);
1152                         }
1153
1154                         kmem_cache_free(blkdev->request_pool, comp_req);
1155                 }
1156
1157                 if (!blkdev->shutting_down) {
1158                         blkvsc_do_pending_reqs(blkdev);
1159                         blk_start_queue(blkdev->gd->queue);
1160                         blkvsc_request(blkdev->gd->queue);
1161                 }
1162         }
1163
1164         spin_unlock_irqrestore(&blkdev->lock, flags);
1165 }
1166
1167 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1168 {
1169         struct blkvsc_request *pend_req, *tmp;
1170         struct blkvsc_request *comp_req, *tmp2;
1171
1172         int ret = 0;
1173
1174         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1175
1176         /* Flush the pending list first */
1177         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1178                                  pend_entry) {
1179                 /*
1180                  * The pend_req could be part of a partially completed
1181                  * request. If so, complete those req first until we
1182                  * hit the pend_req
1183                  */
1184                 list_for_each_entry_safe(comp_req, tmp2,
1185                                          &pend_req->group->blkvsc_req_list,
1186                                          req_entry) {
1187                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1188                                    "sect_start %lu sect_count %ld\n",
1189                                    comp_req,
1190                                    (unsigned long) comp_req->sector_start,
1191                                    comp_req->sector_count);
1192
1193                         if (comp_req == pend_req)
1194                                 break;
1195
1196                         list_del(&comp_req->req_entry);
1197
1198                         if (comp_req->req) {
1199                                 ret = __blk_end_request(comp_req->req,
1200                                         (!comp_req->request.Status ? 0 : -EIO),
1201                                         comp_req->sector_count *
1202                                         blkdev->sector_size);
1203
1204                                 /* FIXME: shouldn't this do more than return? */
1205                                 if (ret)
1206                                         goto out;
1207                         }
1208
1209                         kmem_cache_free(blkdev->request_pool, comp_req);
1210                 }
1211
1212                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1213                            pend_req);
1214
1215                 list_del(&pend_req->pend_entry);
1216
1217                 list_del(&pend_req->req_entry);
1218
1219                 if (comp_req->req) {
1220                         if (!__blk_end_request(pend_req->req, -EIO,
1221                                                pend_req->sector_count *
1222                                                blkdev->sector_size)) {
1223                                 /*
1224                                  * All the sectors have been xferred ie the
1225                                  * request is done
1226                                  */
1227                                 DPRINT_DBG(BLKVSC_DRV,
1228                                            "blkvsc_cancel_pending_reqs() - "
1229                                            "req %p COMPLETED\n", pend_req->req);
1230                                 kmem_cache_free(blkdev->request_pool,
1231                                                 pend_req->group);
1232                         }
1233                 }
1234
1235                 kmem_cache_free(blkdev->request_pool, pend_req);
1236         }
1237
1238 out:
1239         return ret;
1240 }
1241
1242 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1243 {
1244         struct blkvsc_request *pend_req, *tmp;
1245         int ret = 0;
1246
1247         /* Flush the pending list first */
1248         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1249                                  pend_entry) {
1250                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1251                            pend_req);
1252
1253                 ret = blkvsc_submit_request(pend_req,
1254                                             blkvsc_request_completion);
1255                 if (ret != 0)
1256                         break;
1257                 else
1258                         list_del(&pend_req->pend_entry);
1259         }
1260
1261         return ret;
1262 }
1263
1264 static void blkvsc_request(struct request_queue *queue)
1265 {
1266         struct block_device_context *blkdev = NULL;
1267         struct request *req;
1268         int ret = 0;
1269
1270         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1271         while ((req = blk_peek_request(queue)) != NULL) {
1272                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1273
1274                 blkdev = req->rq_disk->private_data;
1275                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1276                     blkdev->media_not_present) {
1277                         __blk_end_request_cur(req, 0);
1278                         continue;
1279                 }
1280
1281                 ret = blkvsc_do_pending_reqs(blkdev);
1282
1283                 if (ret != 0) {
1284                         DPRINT_DBG(BLKVSC_DRV,
1285                                    "- stop queue - pending_list not empty\n");
1286                         blk_stop_queue(queue);
1287                         break;
1288                 }
1289
1290                 blk_start_request(req);
1291
1292                 ret = blkvsc_do_request(blkdev, req);
1293                 if (ret > 0) {
1294                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1295                         blk_stop_queue(queue);
1296                         break;
1297                 } else if (ret < 0) {
1298                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1299                         blk_requeue_request(queue, req);
1300                         blk_stop_queue(queue);
1301                         break;
1302                 }
1303         }
1304 }
1305
1306 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1307 {
1308         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1309
1310         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1311                    blkdev->gd->disk_name);
1312
1313         mutex_lock(&blkvsc_mutex);
1314         spin_lock(&blkdev->lock);
1315
1316         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1317                 spin_unlock(&blkdev->lock);
1318                 check_disk_change(bdev);
1319                 spin_lock(&blkdev->lock);
1320         }
1321
1322         blkdev->users++;
1323
1324         spin_unlock(&blkdev->lock);
1325         mutex_unlock(&blkvsc_mutex);
1326         return 0;
1327 }
1328
1329 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1330 {
1331         struct block_device_context *blkdev = disk->private_data;
1332
1333         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1334                    blkdev->gd->disk_name);
1335
1336         mutex_lock(&blkvsc_mutex);
1337         spin_lock(&blkdev->lock);
1338         if (blkdev->users == 1) {
1339                 spin_unlock(&blkdev->lock);
1340                 blkvsc_do_flush(blkdev);
1341                 spin_lock(&blkdev->lock);
1342         }
1343
1344         blkdev->users--;
1345
1346         spin_unlock(&blkdev->lock);
1347         mutex_unlock(&blkvsc_mutex);
1348         return 0;
1349 }
1350
1351 static int blkvsc_media_changed(struct gendisk *gd)
1352 {
1353         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1354         return 1;
1355 }
1356
1357 static int blkvsc_revalidate_disk(struct gendisk *gd)
1358 {
1359         struct block_device_context *blkdev = gd->private_data;
1360
1361         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1362
1363         if (blkdev->device_type == DVD_TYPE) {
1364                 blkvsc_do_read_capacity(blkdev);
1365                 set_capacity(blkdev->gd, blkdev->capacity *
1366                             (blkdev->sector_size/512));
1367                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1368         }
1369         return 0;
1370 }
1371
1372 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1373 {
1374         sector_t total_sectors = get_capacity(bd->bd_disk);
1375         sector_t cylinder_times_heads = 0;
1376         sector_t temp = 0;
1377
1378         int sectors_per_track = 0;
1379         int heads = 0;
1380         int cylinders = 0;
1381         int rem = 0;
1382
1383         if (total_sectors > (65535 * 16 * 255))
1384                 total_sectors = (65535 * 16 * 255);
1385
1386         if (total_sectors >= (65535 * 16 * 63)) {
1387                 sectors_per_track = 255;
1388                 heads = 16;
1389
1390                 cylinder_times_heads = total_sectors;
1391                 /* sector_div stores the quotient in cylinder_times_heads */
1392                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1393         } else {
1394                 sectors_per_track = 17;
1395
1396                 cylinder_times_heads = total_sectors;
1397                 /* sector_div stores the quotient in cylinder_times_heads */
1398                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1399
1400                 temp = cylinder_times_heads + 1023;
1401                 /* sector_div stores the quotient in temp */
1402                 rem = sector_div(temp, 1024);
1403
1404                 heads = temp;
1405
1406                 if (heads < 4)
1407                         heads = 4;
1408
1409
1410                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1411                         sectors_per_track = 31;
1412                         heads = 16;
1413
1414                         cylinder_times_heads = total_sectors;
1415                         /*
1416                          * sector_div stores the quotient in
1417                          * cylinder_times_heads
1418                          */
1419                         rem = sector_div(cylinder_times_heads,
1420                                          sectors_per_track);
1421                 }
1422
1423                 if (cylinder_times_heads >= (heads * 1024)) {
1424                         sectors_per_track = 63;
1425                         heads = 16;
1426
1427                         cylinder_times_heads = total_sectors;
1428                         /*
1429                          * sector_div stores the quotient in
1430                          * cylinder_times_heads
1431                          */
1432                         rem = sector_div(cylinder_times_heads,
1433                                          sectors_per_track);
1434                 }
1435         }
1436
1437         temp = cylinder_times_heads;
1438         /* sector_div stores the quotient in temp */
1439         rem = sector_div(temp, heads);
1440         cylinders = temp;
1441
1442         hg->heads = heads;
1443         hg->sectors = sectors_per_track;
1444         hg->cylinders = cylinders;
1445
1446         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1447                     sectors_per_track);
1448
1449     return 0;
1450 }
1451
1452 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1453                         unsigned cmd, unsigned long argument)
1454 {
1455 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1456         int ret;
1457
1458         switch (cmd) {
1459         /*
1460          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1461          * than just a GUID. Commented it out for now.
1462          */
1463 #if 0
1464         case HDIO_GET_IDENTITY:
1465                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1466                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1467                                  blkdev->device_id_len))
1468                         ret = -EFAULT;
1469                 break;
1470 #endif
1471         default:
1472                 ret = -EINVAL;
1473                 break;
1474         }
1475
1476         return ret;
1477 }
1478
1479 static int __init blkvsc_init(void)
1480 {
1481         int ret;
1482
1483         BUILD_BUG_ON(sizeof(sector_t) != 8);
1484
1485         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1486
1487         ret = blkvsc_drv_init(BlkVscInitialize);
1488
1489         return ret;
1490 }
1491
1492 static void __exit blkvsc_exit(void)
1493 {
1494         blkvsc_drv_exit();
1495 }
1496
1497 MODULE_LICENSE("GPL");
1498 MODULE_VERSION(HV_DRV_VERSION);
1499 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1500 module_init(blkvsc_init);
1501 module_exit(blkvsc_exit);